first i tried xinit mpv- it showed a fragment of the image that changed every few seconds. then i tried .xinitrc and startx- since then, i’ve been seeing a black screen only. and if before after killing Xorg i fell back into the console, now there is only Malevich’s black square. x11 runs, doesn’t crash, but it’s unclear where it’s displaying. in any case, one frame every few seconds is unacceptable
i began to understand why obsd made its own version of x11 ![]()
Hm what SBC do you use?
On my RPi 3 i can not expect more then 3 FPS I guess ![]()
But even the http stream does not deliver more. I think the problem here is motion delivering mjpeg?
pi2 v1.1. the camera outputs H.264+ 1080p at 25 fps. with hardware acceleration pi2 is quite capable of decoding
Alright, then let’s use the decoder.
For this I downloaded mediamtx, it’s a real-time media server written in go, it can do RTSP and webRTC and some other protocols.
I edited mediamtx.yml and deactivated every protocol except rtsp and set it also to use only tcp
(rtspTransports: [tcp])
Then I streamed my cam via ffmpeg:
ffmpeg -f video4linux2 -framerate 30 -video_size 640x480 -i /dev/video0 -c:v h264_v4l2m2m -b:v 2M -f rtsp -rtsp_transport tcp rtsp://localhost:8554/cam
htop show now an average under 4%, nice!
Then I setup motion again, but I commented out video_device and instead I set netcam_url rtsp://127.0.0.1:8554/cam.
I also reduced the resolution to 640x480 inside motion and again 5 FPS. Should be enough to detect motion ![]()
I also deactivated webcontrol and streaming in motion, we don’t want to watch the janky stream, we only want motion to detect motion and then trigger the script to turn on the monitor
webcontrol_port 0
stream_port 0
htop now shows usage between 13% and 25%, nice!
Motion detection also works, I have a telegram bot running which sends me the videos directly to my phone.
I did some testing, no way you will get 1080p@30fps when motion is also running.
On my machine it’s pretty okay with 1280x720@15fps.
My .xinitrc contains now:
exec mpv --profile=low-latency --untimed rtsp://localhost:8554/cam
And ffmpeg command for streaming:
ffmpeg -f v4l2 -framerate 15 -video_size 1280x720 -i /dev/video0 -vf format=yuv420p -c:v h264_v4l2m2m -b:v 1.5M -g 15 -bf 0 -keyint_min 15 -f rtsp -rtsp_transport tcp rtsp://localhost:8554/cam
hmm.. i don’t understand why you’re doing this. my camera is a separate device with its own processor and operating system, it’s just a black box that continuously transmits a video stream to the network. pi2 accepts it, decodes it(yes, at 1080p 25 fps) and displays it on the connected monitor. the pi2 does not have the task of detecting movement, since the camera does this work, and when it detects movement, it additionally sends messages to the port of the web server located on the pi2. at this point just need to turn on the monitor and turn it off when the messages stop
for serious tasks(as you have outlined), serious solutions are needed, but raspberry pi is garbage that sells at an unreasonably high price(for example: Für GIGABYTE BSRE Mini PC Motherboard mit AMD Ryzen Embedded R1505G/V1605B Dual LAN Soft router Industrielle steuerung computer DIY - AliExpress 502 -It costs half as much as the pi5(even with the 4g ram and 128g ssd), but it wins in every respect. by the way, the chinese sell to you much more high price than to us), it cannot compete with the chinese, it has no future. at the very beginning, there was some optimism and that’s why i bought a raspberry, but for many years i have not found such a garbage task that a raspberry could handle. this is the first one
Ok you never mentioned that. It sounded like you connect a cam directly to the SBC.
I connect just a cheap USB cam to the Pi, basically just a CMOS sensor and nothing else.
Since you already receive probably a RTSP stream or something like that, the situation changes.
And I guess there was a misunderstanding on my side, when you talked about a motion event. I thought you’re talking about https://motion-project.github.io which invokes an event when motion is detected in the stream.
![]()
So you don’t need almost anything of what I said.
Just watch the stream via mvp inside a xsession and then you need some script on your webserver which just executed xset dpms force off and on, when the event is received, right?
IDK, you are comparing apple with bananas. Look at the power consumption of this thing, it needs 65W power supply. (oh no wait, Gigabyte recommends 130W https://www.gigabyte.com/de/Mini-PcBarebone/GB-BSRE-1505-rev-10/sp but to be fair, power consumption should be around 50W)
Also RPi foundation designes whole SBCs, gigabyte just mainboards and then they put all the parts, they buy from other vendors, together. They don’t contribute to kernels, they don’t develop software, they just assemble hardware basically.
The Pi was designed as a educational- and hobbyist platform, with a lot of documentation and community work and software development.
The Rpi is designed for continous 24/7 operation. We’re just abusing it ![]()
And what about the Chinese? Gigabyte is Taiwanese, the CPU on the Mini PC you posted is designed in USA, the GPU also and most of the other components too.
But same for RPi, it’s designed in the UK.
This is not the reason for the price differences, it’s the value creation besides the pure hardware cost.
And good luck with support for the Gigabyte PC. ![]()
this is the maximum consumption, not the constant. the actual consumption depends on the task being performed. my ktm1190adv consumes about five liters of gasoline, but if i open the gas, it will consume 11 liters.
no, the pi was designed to be an affordable computer for the poor. i was present when it was announced ![]()
i completely agree that processors and other complex electronic components are not chinese developments(they’ve only just started making something truly their own). but they are the ones assembling and selling powerful and affordable platforms from these components. and the crucial point is that they do it much cheaper than the original developers- this is the criterion that determines the future.
as for whether Taiwan is part of China, that’s a political question, and each country has its own answer. but all pointing to China soon putting an end to this issue.
as for support, it’s more a question of architecture support than a specific model. and even more abstractly, since my question is about software, not hardware. in the modern world replacing intel platform with amd won’t change anything for the end users(except that they’ll spend less, but get more). and pi is more a problem than an opportunity
right?
yep